"error decoding response body" after upgrading to 0.12.10

[NobodyXu]

https://github.com/cargo-bins/cargo-binstall/actions/runs/12511426634/job/34903489849#step:5:1

ERROR Fatal error:

  × For crate b3sum: Reqwest error: error decoding response body
  ├─▶ Reqwest error: error decoding response body
  ├─▶ error decoding response body
  ╰─▶ there are extra bytes after body has been decompressed

[rjzak]

I think I have a similar situation https://github.com/malwaredb/malwaredb-rs/actions/runs/12510436402/job/34905296392?pr=620: 1: there are extra bytes after body has been decompressed

[NOBLES5E]

It's strange that such a seemingly simple scenario can trigger this "error decoding response body" and that it wasn't caught by the CI tests.

[seanmonstar]

I'll revert and release today.

It'd be nice to understand why the bodies have extra bytes, because this didn't "slip" through CI, the error was designed on purpose. After putting the body in a decompressor and it claims the body is done, reqwest purposefully polls the inner body once to consume the end of a chunked 0\r\n\r\n. If extra bytes are found instead, we assume that's bad. Before this release, that data was just lost. Why are there extra bytes? Are there legitimate uses that should purposely be ignored?

[NobodyXu]

Based on debug logging from cargo-bins/cargo-binstall#2004, the issue happens when downloading from https://index.crates.io/3/b/bat

[seanmonstar]

OK, I think I've found the issue. It was working fine with HTTP/1, but if I connect with HTTP/2, the server sends an extra empty data frame. I'll have a release up that ignores those frames if they are empty.

[seanmonstar]

PR #2508 up with a fix (in case you want to try it before I publish).

[seanmonstar]

v0.12.11 is published with the fix. Sorry for the inconvenience.